Machine Learning-Based Prediction of Student Depression Risk Using the Feature Selection Approach

Abstract

In this study, students' risk of depression was predicted using machine learning classification algorithms. During the data preprocessing stage, categorical variables were digitized, and the data was standardized. The performance of the Decision Tree, Random Forest, Support Vector Machines, Logistic Regression, KNN, and Naive Bayes algorithms were evaluated using accuracy, precision, sensitivity, and F1 score metrics. The results show that Random Forest, Logistic Regression, SVM, and Naive Bayes models stand out with higher F1 scores. Furthermore, a feature selection approach based on Chi-square and correlation was applied, comparing a data set represented by fewer features as a lightweight approach and a data set using all features as a full model approach. The analyses showed that the lite approach produced an F1 score nearly identical to that of the full model. These findings demonstrate that machine learning classifiers perform effectively with the lite approach, which uses fewer features and is fast and feasible for depression risk prediction.